How to Use LiveChat to Qualify Website Leads With an AI Agent

The short answer: if two people answer your website chat and the AI Agent ends up solving about 180 questions a month, Growth is the cheaper LiveChat plan despite its price tag: 2 x $99 = $198 a month with 200 AI resolutions included, against 2 x $25 = $50 on Essential plus 4 packs of 50 extra resolutions for the 170 above the included 10, at $49.50 each, $248 in all (monthly billing, checked on 26 October 2026 on the pricing page). The steps: shape the agent's voice and limits, give it your pricing and product pages, switch on its default skills, build one skill that asks your qualifying questions and opens a lead ticket, test in the preview, then review the runs and the gaps each week.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

New LiveChat trials run on Text, the company's newer product, so the menus below are Text's. The plans are in the LiveChat glossary entry.

Get these right before the first chat:

1. Count resolutions, not chats. For example, 300 chats a month in which the agent answers at least one question in 60% gives 300 x 0.6 = 180 resolutions, the figure behind the sum above. Text counts a resolution when the agent solves at least one question, and not when it hands over without answering (AI Agents article).
2. Your discovery-call questions. The three or four things a salesperson asks first, such as company size, timeline and the job to be done, become the agent's script. Your pricing, product and case-study pages become its knowledge.
3. This month's baseline. Note how many sales conversations started from the website last month and how many sales accepted. That is the comparison after launch, not the agent's own resolution rate.

What to have ready

• A Text account from LiveChat's site: 14 days on Essential or Growth, no card, no automatic charge at the end, and a widget that installs by snippet or plugin (pricing page).
• Owner or Admin rights. Only those roles can edit the AI Agent or switch it on and off (AI Agents article).
• A second team called Sales. Transfers to a chosen team only appear once more than one team exists.

1. Give the agent a voice and its limits

Open the AI Agent's profile from the Team section. Pick a tone (Professional, Concise, Polite, Friendly or Casual) and an answer length, then write the custom voice prompt, which takes up to 10,000 characters, per the AI Agents article. Use it for the limits: never quote a custom price, never promise a delivery date, never give legal advice.

2. Load what a buyer asks about

Under Knowledge sources, add your website. You can take the full site with subpages, only nested pages or a single page, and switch on Auto-update with an interval such as every 30 days. Upload files too, such as a price sheet as PDF (up to 200 pages) or DOCX, and use Articles for answers that need depth, per the AI Agents article.

3. Switch on the default skills

Four of the built-in skills matter for leads, all described in the AI Agents article:

• Send welcome message. Choose the custom, predefined version and put the AI notice in it (see step 6).
• Collect customer details. The agent asks for a name and email when they are missing and saves them to the customer panel; giving them stays optional for the visitor.
• Transfer to human. Choose Transfer to a specific team and pick Sales. Tick Create a ticket, so a lead that waits 30 seconds with nobody free still lands somewhere.
• Handle tickets. Add a prefix such as [AI] so sales can tell the agent's tickets from their own.

4. Build the qualifying skill in Playbook

This is the agent step. Open Playbook, click + Create skill and choose For your AI Agent, as the Playbook article describes. In prompt mode, write the job in plain words: "When a visitor asks about pricing, a demo or a quote, ask for company size, timeline and the main job they need done, then create a ticket with a summary of the answers, tag the chat lead and transfer it to Sales." Check every block Playbook builds before you click Save and activate.

The skill also sees the visitor's details, the page they are on and their last five pages. The agent asks, reads each answer and uses it in the next question; a person on the Sales team reads the ticket and decides whether it is a lead.

If your CRM is Salesforce, add a workspace skill that starts when a ticket is created and uses the dedicated Salesforce actions the builder offers. One AI Agent can carry up to 10 AI Agent skills, per the Playbook article.

5. Rehearse in the preview widget

Ask the preview the 15 questions your sales team hears most. Transfers and ticket creation do not run in preview mode; the widget shows where a skill would have fired instead (AI Agents article). Your sales lead marks each answer right, wrong or missing, and you fix the wrong ones in Knowledge.

6. Add the AI notice and the privacy note

Since 2 August 2026 the EU AI Act's transparency article has applied: a visitor should learn in the first exchange that the other side is an AI system, unless context makes it plain (Article 50, consolidated text on EUR-Lex); the LiveAgent entry has the detail. One line in the welcome message covers it.

A name and email collected in chat are personal data, and Text's privacy note article names GDPR among the reasons for a note, which ships off and empty. Turn it on under Settings > Channels > Website widgets, Customize, then Language > Privacy policy, and link your privacy policy.

7. Go live, then read the runs and the gaps

Switch the agent to Accept chats. In Playbook, the Runs section lists every skill run as Completed, Failed or In Progress, and a failed run shows the exact step that broke (Playbook article). Check it daily for the first week.

Each week, open the agent's Overview and work through Knowledge Gaps: Needs answer holds the questions it could not answer, each with the chats behind it. Write the answer and click Save and train, or Skip the ones that do not matter.

Where Text's articles say to be careful

• The agent makes mistakes. Text's own article ends by telling you to review its chats in the Archives regularly.
• Preview hides the handoff. Test one live chat yourself after launch.
• Cutting packs takes a month. A reduction in resolution packs is a downgrade and only applies at the next renewal (AI Agents article).
• One team means no routing. With a single team, transfers go to the first free teammate, sales or not.

Who pays for the handover

A lead flow that answers a pricing question before it hands over spends a resolution on every lead: Text counts the chat as resolved because one question was solved, although a salesperson then does the real work. That is the definition in the AI Agents article, not a trick, and it means the meter Text bills on climbs with every lead you route. Price the plan on resolutions and judge the agent on leads sales accepts.

What the trial settles

Steps 1 to 5 cost nothing inside the 14 days. A preview that answers your 15 questions the way your strongest salesperson would is the whole test; paying later adds resolutions and seats. A preview that fails points, through Knowledge Gaps and the Runs log, to the piece to fix.

FAQ

Will this work on a legacy LiveChat plan?

No. The pricing FAQ says Text features are not included in LiveChat plans, legacy ones included; you can run a Text trial beside your account (pricing page).

Does the agent count a chat it passes straight to sales?

Not if it answered nothing first. A transfer without an answered question is not a resolution (AI Agents article).

Can I see which chats led to orders?

On a Shopify-connected account, yes: the AI Agent Sales report counts sales during or up to 30 days after an AI-handled chat (AI Agents article). That window credits the chat with every order in the following month, including ones that would have come anyway, so compare it with a week the agent was off.